Q & A with CHDS Programmers

What models are you working on currently?

Stephen Sy: My work has focused on several cervical cancer models that were built previous to my arrival at CHDS. While these are established models, we are constantly updating, upgrading and customizing them for different analyses.  In addition, I’ve started preliminary work on building a coronary heart disease model.

Zachary Ward: I am currently working on two projects: A gastric cancer model, and a model which examines cost-effective approaches to reduce maternal mortality. They are based on existing Markov models but are being coded from scratch, and both are Monte Carlo simulations in which the model records and takes into account what happened to individuals previously in the model.

Can you provide us with a simple walkthrough of the models?

SS: For the main stochastic cervical model, we simulate each individual in a cohort. The model creates an individual woman who is then run through her natural life span month by month. Each month there is a probability that she acquires HPV, which in turn can lead to lesions and then potentially cervical cancer.  If a woman becomes infected, her health state can progress towards cancer or regress towards a more healthy state. Once she reaches the end of her natural life span the model keeps track of her history as well as costs incurred, and this pattern continues until the predetermined cohort size is reached. Throughout the model, interventions can be added, including screening and vaccination. Furthermore, we can customize the model to different countries and/or different populations

ZW: In some ways the maternal and gastric cancer models are similar to the cervical model, but they also differ considerably. Instead of simulating people one at a time, both models create a whole cohort, which are then cycled through the model. One advantage of this is that you can suspend the model whenever you want and take a cross-sectional look at what is occurring at that point in time. We are also planning on running whole populations through the model – basically many birth-cohorts run through the model and stopped at different points, then combined together to create a heterogeneous population with respect to age, which can itself be run through the model.

In the future the maternal model will also incorporate quite a bit of spatial statistics/GIS (geographic information systems) data. Simulated women will have a specific GPS location, as well as attributes that are characteristic of that location, based largely on demographic data.  This increases the amount of heterogeneity we’re able to introduce.  For example, if mortality rates differ between districts, or contraception is more limited in rural locations, we will be able to realistically model these differences.  We will also be able to determine the distance and travel time between a woman’s location and the nearest medical facility.  This opens up new areas of analysis such as looking at how the placement of facilities or road networks could be improved.  Some of the ways we are using the data is quite novel, including how we  link the model to Google Earth, providing us with the ability to create various maps directly from the model.

What type of interface is being utilized?

SS: The cervical model interface is known as CLE (command line entry), there is no user interface analogous to most programs run in Windows. Instead the user enters into a directory at the command prompt and runs the model executable file along with an input file containing all the parameters for the specific setting. For researchers used to this type of interface it is a simple way to program and run the model. This type of interface is quite adaptable, simple to code, and makes it easy to quickly orchestrate changes to input parameters.

ZW: Both the models I have been working are stand-alone applications and are GUI-based (graphical user interface) – they have forms that the user interacts with, like most programs that people use.  Users can import model settings that they have previously defined, such as region-specific population data or age-specific rates of smoking cessation, for example. This type of interface is easier to navigate for people not used to CLE. For the maternal model it will provides us with the potential to deploy the model in a format that is easily accessible to non-specialists.